With the advancement of AI-based traffic prediction and drone technologies, redesigning fixed sensor networks and integrating drone fleets as dynamic supplements presents a cost-effective approach to enhancing traffic management. Traditional sensor networks are typically coverage-driven, aiming to maximize spatial reach. However, recent AI-driven studies have shown that such layouts result in redundant sensors collecting highly correlated data. In practice, reliable predictions can be achieved with fewer sensors, provided they are strategically placed. Motivated by this insight, this study proposes a prediction-driven design for a hybrid monitoring system and explores its relationship with prediction performance.

A two-stage optimization framework is proposed to balance system cost and predictive accuracy. In Stage 1, a genetic algorithm (GA) is used to optimize the infrastructure-level design, minimizing the number of installed fixed sensors, the peak number of simultaneously active drones, and the root mean square error (RMSE) of a pretrained random forest (RF) prediction model. The model is evaluated on masked datasets representing the sensor network’s observable data. In Stage 2, a simulated annealing (SA) algorithm solves a drone routing problem to fulfill the drone monitoring tasks generated in Stage 1 while minimizing operational and procurement costs.

The framework is applied to the pNEUMA dataset, a high-resolution drone-based urban traffic dataset. Results demonstrate that the hybrid system consistently outperforms fixed-sensor-only configurations across all budget levels. The most significant cost savings (up to 40\%) are achieved under moderate budget scenarios. Fixed sensors are shown to be essential for long-term coverage in critical zones with high traffic volume or strong temporal dynamics, while drones effectively supplement under-monitored peripheral regions and provide adaptive temporal monitoring. As the drone fleet size increases, both sensor types are more uniformly distributed, supporting a system design that balances spatial coverage with predictive performance.